Improving Forest Baseline Maps in Tropical Wetlands Using GEDI-Based Forest Height Information and Sentinel-1

Remote Sensing-based global Forest/Non-Forest (FNF) masks have shown large inaccuracies in tropical wetland areas. This limits their applications for deforestation monitoring and alerting in which they are used as a baseline for mapping new deforestation. In radar-based deforestation monitoring, for example, moisture dynamics in unmasked non-forest areas can lead to false detections. We combined a GEDI Forest Height product and Sentinel-1 radar data to improve FNF masks in wetland areas in Gabon using a Random Forest model. The GEDI Forest Height, together with texture metrics derived from Sentinel-1 mean backscatter values, were the most important contributors to the classification. Quantitatively, our mask outperformed existing global FNF masks by increasing the Producer’s Accuracy for the non-forest class by 14%. The GEDI Forest Height product by itself also showed high accuracies but contained Landsat artifacts. Qualitatively, our model was best able to cleanly uncover non-forest areas and mitigate the impact of Landsat artifacts in the GEDI Forest Height product. An advantage of the methodology presented here is that it can be adapted for different application needs by varying the probability threshold of the Random Forest output. This study stresses that, in any application of the suggested methodology, it is important to consider the UA/PA trade-off and the effect it has on the classification. The targeted improvements for wetland forest mapping presented in this paper can help raise the accuracy of tropical deforestation monitoring.

Predicting clinical outcomes in COVID-19 using radiomics on chest radiographs

Objectives For optimal utilization of healthcare resources, there is a critical need for early identification of COVID-19 patients at risk of poor prognosis as defined by the need for intensive unit care and mechanical ventilation. We tested the feasibility of chest X-ray (CXR)-based radiomics metrics to develop machine-learning algorithms for predicting patients with poor outcomes. Methods In this Institutional Review Board (IRB) approved, Health Insurance Portability and Accountability Act (HIPAA) compliant, retrospective study, we evaluated CXRs performed around the time of admission from 167 COVID-19 patients. Of the 167 patients, 68 (40.72%) required intensive care during their stay, 45 (26.95%) required intubation, and 25 (14.97%) died. Lung opacities were manually segmented using ITK-SNAP (open-source software). CaPTk (open-source software) was used to perform 2D radiomics analysis. Results Of all the algorithms considered, the AdaBoost classifier performed the best with AUC = 0.72 to predict the need for intubation, AUC = 0.71 to predict death, and AUC = 0.61 to predict the need for admission to the intensive care unit (ICU). AdaBoost had similar performance with ElasticNet in predicting the need for admission to ICU. Analysis of the key radiomic metrics that drive model prediction and performance showed the importance of first-order texture metrics compared to other radiomics panel metrics. Using a Venn-diagram analysis, two first-order texture metrics and one second-order texture metric that consistently played an important role in driving model performance in all three outcome predictions were identified. Conclusions: Considering the quantitative nature and reliability of radiomic metrics, they can be used prospectively as prognostic markers to individualize treatment plans for COVID-19 patients and also assist with healthcare resource management. Advances in knowledge We report on the performance of CXR-based imaging metrics extracted from RT-PCR positive COVID-19 patients at admission to develop machine-learning algorithms for predicting the need for ICU, the need for intubation, and mortality, respectively.

An Examination of an Enhanced Remote Sensing Method for Agent Attribution of Forest Disturbance

Patterns of disturbance in Sierra Nevada forests are shifting as a result of changing climate and land uses. These changes have underscored the need for a monitoring system that both detects disturbances and attributes them to different agents. Addressing this need will aid forest management and conservation decision-making, potentially enhancing forests’ resilience to changing climatic conditions. In addition, it will advance understanding of the patterns, drivers, and consequences of forest disturbance in space and time. This study proposed and evaluated an enhanced method for disturbance agent attribution. Specifically, it tested the extent to which textural information could improve the performance of an ensemble learning method in predicting the agents of disturbance from remote sensing observations. Random Forest (RF) models were developed to attribute disturbance to three primary agents (fire, harvest, and drought) in Stanislaus National Forest, California, U.S.A., between 1999 and 2015. To account for spectral behavior and topographical characteristics that regulate vegetation and disturbance dynamics, the models were trained on predictors derived from both the Landsat record and from a digital elevation model. The predictors included measurements of spectral change acquired through temporal segmentation of Landsat data; measurements of patch geometry; and a series of landscape texture metrics. The texture metrics were generated using the Grey-Level Co-Occurrence Matrix (GLCM). Two models were produced: one with GLCM texture metrics and one without. The per-class and overall accuracies of each model were evaluated with out-of-bag (OOB) observations and compared statistically to quantify the contribution of texture metrics to classification skill. Overall OOB accuracy was 72.0% for the texture-free model and 72.2% for the texture-dependent model, with no significant accuracy difference between them. Spatial patterns in prediction maps cohered with expectations, with most harvest concentrated in mid-elevation forests and fire and stress co-occurring at lower elevations. Altogether, the method yielded adequate identification of disturbance and moderate attribution accuracy for multiple disturbance agents. While textures did not contribute meaningfully to model skill, the study offers a strong foundation for future development, which should focus on improving the efficacy of the model and generalizing it for systems beyond the Central Sierra Nevada.

Mapping Tree Species Deciduousness of Tropical Dry Forests Combining Reflectance, Spectral Unmixing, and Texture Data from High-Resolution Imagery

In tropical dry forests, deciduousness (i.e., leaf shedding during the dry season) is an important adaptation of plants to cope with water limitation, which helps trees adjust to seasonal drought. Deciduousness is also a critical factor determining the timing and duration of carbon fixation rates, and affecting energy, water, and carbon balance. Therefore, quantifying deciduousness is vital to understand important ecosystem processes in tropical dry forests. The aim of this study was to map tree species deciduousness in three types of tropical dry forests along a precipitation gradient in the Yucatan Peninsula using Sentinel-2 imagery. We propose an approach that combines reflectance of visible and near-infrared bands, normalized difference vegetation index (NDVI), spectral unmixing deciduous fraction, and several texture metrics to estimate the spatial distribution of tree species deciduousness. Deciduousness in the study area was highly variable and decreased along the precipitation gradient, while the spatial variation in deciduousness among sites followed an inverse pattern, ranging from 91.5 to 43.3% and from 3.4 to 9.4% respectively from the northwest to the southeast of the peninsula. Most of the variation in deciduousness was predicted jointly by spectral variables and texture metrics, but texture metrics had a higher exclusive contribution. Moreover, including texture metrics as independent variables increased the variance of deciduousness explained by the models from R2 = 0.56 to R2 = 0.60 and the root mean square error (RMSE) was reduced from 16.9% to 16.2%. We present the first spatially continuous deciduousness map of the three most important vegetation types in the Yucatan Peninsula using high-resolution imagery.

Assessment of Texture Features for Bermudagrass (Cynodon dactylon) Detection in Sugarcane Plantations

Sugarcane products contribute significantly to the Brazilian economy, generating U.S. $12.2 billion in revenue in 2018. Identifying and monitoring factors that induce yield reduction, such as weed occurrence, is thus imperative. The detection of Bermudagrass in sugarcane crops using remote sensing data, however, is a challenge considering their spectral similarity. To overcome this limitation, this paper aims to explore the potential of texture features derived from images acquired by an optical sensor onboard anunmanned aerial vehicle (UAV) to detect Bermudagrass in sugarcane. Aerial images with a spatial resolution of 2 cm were acquired from a sugarcane field in Brazil. The Green-Red Vegetation Index and several texture metrics derived from the gray-level co-occurrence matrix were calculated to perform an automatic classification using arandom forest algorithm. Adding texture metrics to the classification process improved the overall accuracy from 83.00% to 92.54%, and this improvement was greater considering larger window sizes, since they representeda texture transition between two targets. Production losses induced by Bermudagrass presence reached 12.1 tons × ha−1 in the study site. This study not only demonstrated the capacity of UAV images to overcome the well-known limitation of detecting Bermudagrass in sugarcane crops, but also highlighted the importance of texture for high-accuracy quantification of weed invasion in sugarcane crops.